Documentation / Calibration / Out-of-tolerance events

I — The event

A gage fails calibration. Which parts are suspect?

An out-of-tolerance event is an instrument arriving for calibration whose as-found condition is outside its acceptance limits. Its measurements are then unreliable back to some unknown point, so the standard assumption is that it was out since its last known-good calibration and the entire preceding interval is suspect.

Reference · No product claim on this page
“As-found” is the load-bearing word. If the calibration lab adjusts the instrument first and reports only as-left data, the event is invisible and the evidence is gone. As-found values are something to require on the purchase order and verify on receipt rather than to assume.

II — The exposure window

The window runs from the last known-good calibration to the moment the instrument left service. Everything measured inside it is in scope.

That is also where the money is. If intermediate checks exist, the window shortens to the last passing check, which is the single largest economic argument for running them, and it is an argument almost nobody makes when the checks are proposed. A quarterly instrument with a weekly check has a week of exposure instead of a quarter.

last known-good calibration
  → ... measurements ...
  → last passing intermediate check      ← the window can start here instead
  → ... measurements ...
  → removed from service / found OOT

III — The impact analysis

  1. Bound the window. Last known-good calibration to removal from service, shortened by any passing intermediate check.
  2. Enumerate what was measured in it: which instrument measured which characteristic, on which part, on which date. This is the step that fails.
  3. Quantify the error, with magnitude and sign, at the relevant point of the instrument’s range.
  4. Assess against the product tolerance. The question is not “was the gage out of tolerance” but “could its error have flipped a decision”.
  5. Assess the reverse impact. An out-of-tolerance gage may also have rejected good product.
  6. Decide and document, including a finding of nil impact, which has to be recorded rather than assumed.
  7. Correct the root cause, which is rarely the instrument.
Sign is not a detail. A gage reading low against an upper limit accepts oversized parts. The same gage reading high rejects good ones. Opposite failures, opposite costs, and the sign is frequently the thing the calibration record does not carry.

IV — Why organizations under-quarantine

Quarantine is a decision under uncertainty with asymmetric costs, usually taken by a quality manager with incomplete data and a schedule pressing on them. What legitimately bears on it is the magnitude of the error against tolerance, its direction, the distribution of the values actually measured, whether anything downstream would have caught the escape, and the traceability of the product itself.

That last one dominates everything else. Where the affected lots cannot be identified, the decision collapses to “everything since the last calibration”, and organizations under schedule pressure then systematically under-quarantine. The mechanism is worth naming precisely: it is an inability to identify the affected product, followed by a rationalisation, and it presents from outside as a decision to accept risk.

V — What the records system must already hold

Reverse traceability is a data-model problem rather than a metrology problem. Whether the analysis is possible at all was decided at the moment each measurement was taken, and none of it can be reconstructed afterwards.

Stored at measurement timeWhy it is load-bearing
The gage instance ID rather than the gage typeThe difference between a twenty-minute analysis and a plant-wide recall. Most records say “inspected by J. Smith”, not “measured with MIC-0147”.
A timestampTo place the measurement inside the window.
Product identity at the finest granularity availableSerial, then lot, then work order, then shift. Determines what can be quarantined selectively.
The characteristic and its toleranceNeeded to ask whether the error could have flipped the decision.
The measured value rather than pass/fail alonePass/fail-only records make impact analysis unresolvable and force worst-case quarantine.
As-found values per check point, numerically, with signMagnitude and direction both come from here.
The acceptance limits that appliedSo “out of tolerance” has a definition as of the date it applied.
Which reference standards were used, by IDNeeded for cascading recall.
Issue and return historyBounds which measurements the gage could plausibly have made, shortening the window.
Intermediate check recordsThe single largest cost lever in the domain.
An immutable, timestamped audit trail over all of itNeeded for the analysis to be credible to an auditor.
The cascade is the case nobody plans for. If a reference standard is found out of tolerance, every calibration performed against it since its own last known-good result is in question, and so is every instrument those calibrations passed.

VI — Where this connects to measurement capture

Step two is the one that fails, and it fails for a reason that sits below the records system entirely. The wire protocol under most hand gaging carries a value, a sign, a decimal position and a unit and nothing else: no instrument identifier at all. Identity is manufactured above the gage, at the port or in software, which means it is only present in the record if something bound it there at the time.

That is why the first table above starts with instance ID rather than ending with it. A records system cannot add that field retrospectively, however good it is; it can only ever store what capture handed it.

VII — Sources and boundaries